Research map / Image generation / Text-to-image diffusion
AttriCtrl: A Generalizable Framework for Controlling Semantic Attribute Intensity in Diffusion Models
Die Chen, Zhongjie Duan, Zhiwen Li, Cen Chen, Daoyuan Chen, Yaliang Li, Yinda Chen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add positional encoding to value encoder token sequence (ablation) | AttriCtrl · GenEval brightness control | AvgDiff: 0.181 -> 0.141 | Improved |
| add positional encoding to value encoder token sequence (ablation) | AttriCtrl · GenEval realism control | AvgDiff: 0.228 -> 0.192 | Improved |
| add positional encoding to value encoder token sequence (ablation) | AttriCtrl · GenEval detail control | AvgDiff: 0.213 -> 0.191 | Improved |
| use 32 tokens instead of 1 token (ablation) | AttriCtrl · GenEval brightness control | AvgDiff: 0.257 -> 0.141 | Improved |
| use 32 tokens instead of 8 tokens (ablation) | AttriCtrl · GenEval brightness control | AvgDiff: 0.185 -> 0.141 | Improved |
| use 32 tokens instead of 16 tokens (ablation) | AttriCtrl · GenEval detail control | AvgDiff: 0.253 -> 0.191 | Improved |
| increase tokens from 32 to 64 (ablation) | AttriCtrl · GenEval brightness control | AvgDiff: 0.141 -> 0.171 | No clear effect |
| increase tokens from 32 to 64 (ablation) | AttriCtrl · GenEval realism control | AvgDiff: 0.192 -> 0.196 | No clear effect |
| learned value encoder instead of interpolation (AID-in) | AttriCtrl vs AID-in · GenEval attribute control (average) | AvgDiff: 0.267 -> 0.175 | Improved |
| learned value encoder instead of prompt instruction (Kontext) | AttriCtrl vs Kontext · GenEval attribute control (average) | AvgDiff: 0.328 -> 0.175 | Improved |
| learned value encoder instead of two-endpoint weighted embedding (W-Emb) | AttriCtrl vs W-Emb · GenEval attribute control (average) | AvgDiff: 0.345 -> 0.175 | Improved |
| safety value encoder instead of concept erasure ESD | AttriCtrl vs ESD · I2P | removal rate (%): 53.9 -> 57.7 | Improved |
Closest papers by meaning
- Aesthetic Post-Training Diffusion Models from Generic Preferences with Step-by-step Preference Optimization (CVPR 2025 · similarity 0.89)
- Unlocking the Essence of Beauty: Advanced Aesthetic Reasoning with Relative-Absolute Policy Optimization (ICLR 2026 · similarity 0.89)
- Bridging Cognitive Gap: Hierarchical Description Learning for Artistic Image Aesthetics Assessment (AAAI 2026 · similarity 0.88)
- PrefPaint: Aligning Image Inpainting Diffusion Model with Human Preference (NeurIPS 2024 · similarity 0.88)
- Aligning Vision Models with Human Aesthetics in Retrieval: Benchmarks and Algorithms (NeurIPS 2024 · similarity 0.88)
- Style Injection in Diffusion: A Training-free Approach for Adapting Large-scale Diffusion Models for Style Transfer (CVPR 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
